How an AI video is actually made
Four stages, and only one of them is the one people picture. Here is what each needs from you, what it costs in time, and where it goes wrong.
The four stages
-
Script
One sentence about what the video has to achieve, turned into shots. This is where a project is won or lost: a shot list that ignores what models are bad at will cost a week in the middle.
-
Generation
Each shot generated against the agreed look, usually several times. The first pass is about finding the ones worth keeping, not about finishing anything.
-
Selection
The part nobody advertises and the part that takes longest. For every shot that survives, several are thrown away — hands, text, continuity between cuts.
-
Edit
Cut, grade, sound, captions and typography. Logo work and any text on screen is done here rather than generated, because this is where accuracy matters.
Where the time actually goes
People assume generation is the slow part. It is not — it is selection. A model will give you ten versions of a shot in the time it takes to read this paragraph, and nine of them will have something wrong with a hand, a reflection or a word on a label. Someone has to watch all ten.
That ratio is the whole difference between a finished piece and a demo, and it is why a three-day turnaround is honest for one ad and dishonest for a brand film. What you are buying is the discarding.
What the discarding was done with: the tools behind the work. What survived it: the work.
Questions about the process
How long does one AI video take?
Three to five working days for a single ad from an approved script; seven to ten for a longer brand film. Most of that time is selection rather than generation — for every shot that ends up in the cut, several are thrown away.
What do I need to have ready before we start?
One sentence about what the video has to achieve and who it is for. A brand book or palette, product photography, previous videos you liked and a deadline all help. If there is no script, we write it — that is the first step of the process, not a prerequisite.
How many rounds of revisions are included?
Two: one after the first version and one after the edit. Each round collects every comment into a single list, because comments arriving one at a time is what delays projects more than anything else. Further rounds are available and billed separately.
What formats do I get?
MP4 (H.264) in whichever aspect ratios you need — 9:16, 1:1, 16:9. A separate subtitle file comes with it, and a clean version without text overlays where you plan to run it in other languages. Higher-quality masters on request.
Can you work from a script we already have?
Yes. We start from it and talk through which shots are realistically generatable. Some things that are trivial to film are hard for a model, and the reverse is just as true — that usually means a few changes to the framing rather than a rewrite.